Multiple Synthetic ECG Image Datasets for Digitization, Lead region and lead name Detection, and Signal Segmentation

ECG Image Dataset for Digitization, Detection, and Segmentation Multiple ECG datasets for deep learning-based tasks, including ECG digitization, lead region and lead name detection, and pixel-wise waveform segmentation. A detailed description is provided in our paper. You can also use our open-source Python framework to generate large-scale, customizable ECG image datasets. 📄 Paper: An Open-Source Python Framework and Synthetic ECG Image Datasets for Digitization, Lead and Lead Name Detection, and Overlapping Waveform Segmentation (To appear) 🔗 Paper at Journal of Medical Signals & Sensors SynthECG: Python Framework and ECG Image Datasets for Digitization, Lead Detection, and Waveform Segmentation 🔗 Python framework: https://github.com/rezakarbasi/ecg-image-and-signal-dataset 🔗 Arxiv link of the paper: https://doi.org/10.48550/arXiv.2506.06315 📢 Citation If you use this dataset or code, please cite both the dataset and the paper: Paper: Rahimi, Masoud*; Karbasi, Reza*; Vahabie, Abdol-Hossein. SynthECG: Python Framework and ECG Image Datasets for Digitization, Lead Detection, and Waveform Segmentation. Journal of Medical Signals & Sensors 16(3):8, March 2026. | DOI: 10.4103/jmss.jmss_58_25 Dataset: Rahimi, M., Karbasi, R., & Vahabie, A. H. (2025). An Open-Source Python Framework and Synthetic ECG Image Datasets for Digitization, Lead and Lead Name Detection, and Overlapping Signal Segmentation. University of Tehran. 📊 Dataset Summary Task Format/Annotations Sample Size ECG Digitization ECG images + ground truth signals 2000 Lead & Lead Name Detection ECG images + Bounding boxes (YOLO format) 2000 Segmentation (Normal) single leads + masks + ground truth signals 20000 Segmentation (Overlapping) Overlapping leads + clean binary masks 100(with operlap), 185(non-overlap) 📦 Key Features Realistic ECG image and ground truth signals generated in multiple layouts: 3x1, 3x4, 6x2, 12x1 YOLO-format bounding boxes for lead and lead name detection Pixel-level segmentation masks compatible with U-Net models Support for overlapping signal segmentation with clean, non-overlapping masks Paired time-series ground truth signals for all segmentation samples